Security

Control who sees what — and prove it later.

Generic AI assistants are fast but untrustworthy for operations — they can hallucinate a result, make a different decision on the same input, and leave no audit trail. Atagente is built for the messy middle: enough AI to remove the drudgery, enough determinism and oversight to trust the outcome.

Architecture — request path

Every input travels the same guarded path.

Eight controls between an untrusted input and an executed action. Select a layer to see what it does.

LAYER 05AVAILABLE

The decision is code

The judgment stays with AI; the decision is computed in code. A gate that says escalate over $200 will route identically every single time — testable and provable.

guaranteesame input → same answer
branchesapprove / escalate / reject
recordthe rule and the outcome are logged
Controls

Why should I trust it?

Credentials are encrypted and injected at call time, AI routes are rate-limited, records are tenant-tagged, and governance and audit are first-class.

01MVP

Deterministic decision gates

Decisions computed in code; the same input always gives the same answer.

02MVP

Fail-loud handling

Errors surface or hand off to a human, never a fabricated result.

03MVP

Injection guard

User-supplied content is treated as data, never as instructions to the AI.

04MVP

Idempotency

Stable keys on external actions so a retry can’t run the same action twice.

05MVP

Encrypted credential store

Secrets held in an encrypted store with rotation, and injected at call time.

06MVP

Rate-limited AI routes

Model-facing routes are rate limited so no tenant or run can flood them.

07MVP

Multi-tenant record tagging

Every record is tagged to its tenant, so data never crosses an organisation boundary.

08MVP

Governance registry

A registry for ownership and naming, so every workflow and connector has an owner.

09MVP

Observability with auto-recovery

Runtime health is observable, and interrupted runs recover from their last checkpoint.

Access model

Internal and external, cleanly separated.

Authenticated access + OTP

Every login, internal and portal, is verified — no shared credentials.

Scoped external access

Stakeholders only ever see their own case, never the wider system.

Full audit logging

Every status change, approval, and output is logged automatically for later review.

Sandboxed automation

Generated automation runs isolated, without direct access to credentials or unrelated data.

PLANNED — NOT AVAILABLE TODAY

Governance. Role-based access control, a per-step audit trail, and start-with-payload inputs.

Resilience. Timeouts and cost caps on every model call, compensating transactions, and scale-out of workers.

Enterprise. SSO / SCIM, per-tenant limits, exportable and tamper-evident audit, and managed backups.

Built for the teams who have to sign off.

Bring your security questions to the demo — we'll walk the request path with you.

Book a DemoRead the FAQ